ProgramBench task detail

junegunn__fzf.b56d614

Task-level results for this Codex /goal scaffold. ProgramBench baseline context is cached from the official task page; Codex scored tests are after active-branch and ignored-test filtering.

1858generated behavioral tests
82.4%official best score
82.1%Codex best score
3Codex result rows

Codex Results by Model

#ModelRunAgentModeScoreEvalTestsEst. costCallsWallEvidence
1 GPT 5.5 (xhigh) 20260517T094700Z Codex /goal Mini-SWE-compatible no internet 82.1% ok 1526/1858 $6.12 90 0.27h manifest · eval json · eval summary · usage audit
2 GPT 5.5 (xhigh) 20260518T232006Z Codex /goal Paper prompt + /goal no internet 80.4% ok 1494/1858 $4.31 50 0.22h manifest · eval json · eval summary · agent summary · usage audit
3 GPT 5.5 (xhigh) 20260519T215046Z Codex /goal Paper prompt + Goal contract no internet 79.2% ok 1472/1858 $3.67 49 0.21h not exported

Why Scores Differ

Official ProgramBench rows are the public mini-SWE-agent submissions. GoalBench rows are separate Codex /goal submissions. Same model label does not mean the same agent, prompt, tool loop, or generated implementation. A GoalBench row is resolved only when the ProgramBench-scored pass rate is exactly 100%.

GPT 5.5 (xhigh) · 20260517T094700Z

82.1% from 1526/1858 ProgramBench-scored tests. Raw public eval statuses: failure: 445, passed: 1711, skipped: 8.

Agent trace summary unavailable.

Likely miss class: behavioral mismatch, interactive terminal behavior mismatch.

  • e31996014aaf eval.tests.test_basic.test_tail_limit (failure)
  • e31996014aaf eval.tests.test_errors.test_history_size_zero (failure)
  • e31996014aaf eval.tests.test_errors.test_height_over_100_percent (failure)
  • e31996014aaf eval.tests.test_interactive_tmux.test_interactive_print_query (failure)
  • e31996014aaf eval.tests.test_errors.test_tail_zero (failure)
  • e31996014aaf eval.tests.test_interactive_tmux.test_interactive_selection (failure)

manifest · eval json · eval summary · usage audit

GPT 5.5 (xhigh) · 20260518T232006Z

80.4% from 1494/1858 ProgramBench-scored tests. Raw public eval statuses: failure: 482, passed: 1674, skipped: 8.

Agent trace: 134 shell command(s), 93 executable-observation command(s), 9 build command(s), 3 blocked attempt(s). Raw Codex logs are not published.

Likely miss class: behavioral mismatch, interactive terminal behavior mismatch.

  • e31996014aaf eval.tests.test_errors.test_height_over_100_percent (failure)
  • e31996014aaf eval.tests.test_errors.test_tail_zero (failure)
  • 6e634b67e0e9 eval.tests.test_basic_invocation.test_bash_integration_is_executable_bash (failure)
  • 6e634b67e0e9 eval.tests.test_advanced_options.test_tail_zero (failure)
  • 6e634b67e0e9 eval.tests.test_basic_invocation.test_bash_integration_script (failure)
  • 6e634b67e0e9 eval.tests.test_error_handling.test_invalid_walker_option (failure)

manifest · eval json · eval summary · agent summary · usage audit

GPT 5.5 (xhigh) · 20260519T215046Z

Public eval evidence has not been exported for this row yet.

Official ProgramBench Results by Model

#ModelProviderScoreCostCalls
1 GPT 5.5 (xhigh) OpenAI 82.4% $8.19 91
2 Claude Opus 4.6 Anthropic 81.9% $11.78 253
3 Claude Sonnet 4.6 Anthropic 81.1% $29.33 607
4 GPT 5.5 (high) OpenAI 79.2% $3.76 52
5 GPT 5.5 OpenAI 77.6% $0.97 17
6 Claude Opus 4.7 (xhigh) Anthropic 77.2% $1.80 53
7 Claude Opus 4.7 Anthropic 74.9% $0.56 20
8 GPT 5.4 OpenAI 71.9% $0.34 12
9 Gemini 3.1 Pro Google 62.8% $0.77 55
10 GPT 5.4 mini OpenAI 46.6% $0.04 7
11 Claude Haiku 4.5 Anthropic 22.3% $0.72 105
12 Gemini 3 Flash Google 13.9% $0.16 45
13 GPT 5 mini OpenAI 3.7% $0.03 12